A method for generating a data sample in a first frequency band from measurements in a second frequency band. The method includes obtaining a first plurality of samples, obtaining a second plurality of samples, obtaining a mapping model based on the first plurality of samples and the second plurality of samples, obtaining a third plurality of samples, and obtaining the data sample based on the mapping model and the third plurality of samples. Obtaining the first plurality of samples includes measuring a first frequency response of an environment in the first frequency band. Obtaining the second plurality of samples includes measuring a second frequency response of the environment in the second frequency band. Obtaining the third plurality of samples includes measuring a third frequency response of the environment in the second frequency band. Obtaining the data sample includes applying the mapping model on the third plurality of samples.
Legal claims defining the scope of protection. Each claim is shown in both the original legal language and a plain English translation.
11. The method of claim 8, wherein applying the ResNet comprises applying a plurality of ResNet blocks by extracting the third plurality of feature maps from an output of an Lrth ResNet block of the plurality of ResNet blocks where Lr is a number of the plurality of ResNet blocks, extracting the third plurality of feature maps comprising obtaining an (lr+1)th plurality of residual feature maps by applying an lrth ResNet block of the plurality of ResNet blocks on an lrth plurality of residual feature maps where 1≤lr≤Lr, a first plurality of residual feature maps comprising the second plurality of feature maps and the lrth ResNet block comprising two cascaded convolutional layers and a residual connection.
This invention relates to deep learning-based image processing, specifically improving feature extraction using ResNet (Residual Network) architectures. The problem addressed is enhancing the accuracy and efficiency of feature extraction in convolutional neural networks (CNNs) by leveraging residual connections and cascaded convolutional layers. The method involves applying a ResNet architecture to extract feature maps from an input image. The ResNet consists of multiple ResNet blocks, each containing two cascaded convolutional layers and a residual connection. The residual connection ensures that the input to each block is added to its output, mitigating the vanishing gradient problem and improving training efficiency. The feature extraction process starts with an initial set of feature maps, which are processed sequentially through each ResNet block. Each block generates a new set of residual feature maps by applying its convolutional layers to the previous block's output. The final output is a refined set of feature maps, optimized for downstream tasks such as classification or object detection. The number of ResNet blocks (Lr) can be adjusted based on the complexity of the task, with each block contributing to progressively deeper feature extraction. This approach enhances the network's ability to capture hierarchical features while maintaining computational efficiency.
Cooperative Patent Classification codes for this invention. Click any code to explore related patents in that topic.
October 25, 2021
June 11, 2024
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.